ProDeo 3.1 - The PESTO version

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
Rebel
Posts: 5803
Joined: Thu Aug 18, 2011 10:04 am

ProDeo 3.1 - The PESTO version

Post by Rebel » Fri Feb 19, 2021 8:38 pm

PESTO stands for Piece Square Tables Only as described on the Rofchade website by Dutchman Ronald Friederich and later implemented in TSCP by Rodent author Pawel Koziol which gave TSCP an 200 elo boast.

Inspired by this I implemented PESTO in ProDeo (on top of the existing PST's) just for the fun of it and for a 3 days work (one day programming, two days testing) the results (+50 elo) are very good for ProDeo. I therefore can recommend every engine programmer to give it the same try, hence this release.
....

http://rebel13.nl/prodeo/prodeo-3.1.html
90% of coding is debugging, the other 10% is writing bugs.

User avatar
pedrox
Posts: 1010
Joined: Fri Mar 10, 2006 5:07 am
Location: Basque Country (Spain)
Contact:

Re: ProDeo 3.1 - The PESTO version

Post by pedrox » Sat Feb 20, 2021 1:45 am

I use the evaluation not only on terminal nodes but also on internal nodes as using only material more margin for pruning does not give me good results.

I have tried to use PeSTO evaluation on each internal node for pruning and the result has not been conclusive, but maybe I don't lose Elo and maybe I gain +5. What I notice is that my engine does not seem faster even though the evaluation is much smaller.

But when I replace my evaluation by PeSTO evaluation I lose about 150 Elo points.

I don't know why this works in Prodeo and not in my engine. I thought that one of the weak parts of my engine was the evaluation, I supposed it was worse than Prodeo and and therefore it seems that it should work easier. So I'm a bit confused.

User avatar
pedrox
Posts: 1010
Joined: Fri Mar 10, 2006 5:07 am
Location: Basque Country (Spain)
Contact:

Re: ProDeo 3.1 - The PESTO version

Post by pedrox » Sat Feb 20, 2021 1:53 am

It would be interesting to see a Fruit 2.1 with PeSTO evaluation.

JohnW
Posts: 246
Joined: Wed Nov 21, 2012 11:20 pm
Location: New Hampshire

Re: ProDeo 3.1 - The PESTO version

Post by JohnW » Sat Feb 20, 2021 2:28 am

Played a few games between 3.1 and 2.0 @ 1+1 time controls, seems pretty close..

ProDeo 3.1 +5/=11/-4
ProDeo 2.0 +4/=11/-5

User avatar
Rebel
Posts: 5803
Joined: Thu Aug 18, 2011 10:04 am

Re: ProDeo 3.1 - The PESTO version

Post by Rebel » Sat Feb 20, 2021 8:23 am

pedrox wrote:
Sat Feb 20, 2021 1:45 am
I use the evaluation not only on terminal nodes but also on internal nodes as using only material more margin for pruning does not give me good results.

I have tried to use PeSTO evaluation on each internal node for pruning and the result has not been conclusive, but maybe I don't lose Elo and maybe I gain +5. What I notice is that my engine does not seem faster even though the evaluation is much smaller.

But when I replace my evaluation by PeSTO evaluation I lose about 150 Elo points.

I don't know why this works in Prodeo and not in my engine. I thought that one of the weak parts of my engine was the evaluation, I supposed it was worse than Prodeo and and therefore it seems that it should work easier. So I'm a bit confused.
It's a known fact what works for engine X doesn't have to work for engine Y and vice versa. 4 points comes to mind -

1. I used Pawel's source code for TSCP as found on the TSCP page, pesto.zip.
2. I used the PESTO PST's without the piece values else there will be a conflict with your own piece values.
3. I used the PESTO PST's on top of the PST evaluation in ProDeo, there is some other good stuff in it that I did not want to replace (lose).
4. I noticed the main elo gain came from the evaluation (+45) and little from the search changes, +5 max.
90% of coding is debugging, the other 10% is writing bugs.

User avatar
hgm
Posts: 25860
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller
Contact:

Re: ProDeo 3.1 - The PESTO version

Post by hgm » Sat Feb 20, 2021 11:19 am

I am not sure what needs to be implemented. Most engines already have Piece-Square tables, often in addition to other evaluation terms, so it seems this is just a matter of tuning the eval parameters they already have. (In particular, setting the weight for all other terms to zero, and tune all the PST values.)

Of course just copying the values from an engine that has optimized them, and where they work excessively well, saves a lot of tuning effort. But the problem is of course that it isn't your own engine anymore. Micro-Max would probably also be 3000+ Elo if I borrowed Stockfish' search and Leela's eval.

User avatar
pedrox
Posts: 1010
Joined: Fri Mar 10, 2006 5:07 am
Location: Basque Country (Spain)
Contact:

Re: ProDeo 3.1 - The PESTO version

Post by pedrox » Sat Feb 20, 2021 7:33 pm

Rebel wrote:
Sat Feb 20, 2021 8:23 am
pedrox wrote:
Sat Feb 20, 2021 1:45 am
I use the evaluation not only on terminal nodes but also on internal nodes as using only material more margin for pruning does not give me good results.

I have tried to use PeSTO evaluation on each internal node for pruning and the result has not been conclusive, but maybe I don't lose Elo and maybe I gain +5. What I notice is that my engine does not seem faster even though the evaluation is much smaller.

But when I replace my evaluation by PeSTO evaluation I lose about 150 Elo points.

I don't know why this works in Prodeo and not in my engine. I thought that one of the weak parts of my engine was the evaluation, I supposed it was worse than Prodeo and and therefore it seems that it should work easier. So I'm a bit confused.
It's a known fact what works for engine X doesn't have to work for engine Y and vice versa. 4 points comes to mind -

1. I used Pawel's source code for TSCP as found on the TSCP page, pesto.zip.
2. I used the PESTO PST's without the piece values else there will be a conflict with your own piece values.
3. I used the PESTO PST's on top of the PST evaluation in ProDeo, there is some other good stuff in it that I did not want to replace (lose).
4. I noticed the main elo gain came from the evaluation (+45) and little from the search changes, +5 max.
I used the first option. I have tested PeSTO on a TSCP-derived engine (Caligula), so it is even possible to copy the TSCP evaluation directly with copy-paste.

But I understand now, that in ProDeo not all your evaluation has been replaced by PeSTO from TSCP, I understand that you keep things from the evaluation like maybe mobility, king safety ...

The interesting thing about PeSTO if I'm not confused, is that its author with this evaluation (material + pst) already has 3100 Elo points.

https://ccrl.chessdom.com/ccrl/404/cgi/ ... 4-bit_4CPU

User avatar
Rebel
Posts: 5803
Joined: Thu Aug 18, 2011 10:04 am

Re: ProDeo 3.1 - The PESTO version

Post by Rebel » Sat Feb 20, 2021 8:18 pm

pedrox wrote:
Sat Feb 20, 2021 7:33 pm
Rebel wrote:
Sat Feb 20, 2021 8:23 am
pedrox wrote:
Sat Feb 20, 2021 1:45 am
I use the evaluation not only on terminal nodes but also on internal nodes as using only material more margin for pruning does not give me good results.

I have tried to use PeSTO evaluation on each internal node for pruning and the result has not been conclusive, but maybe I don't lose Elo and maybe I gain +5. What I notice is that my engine does not seem faster even though the evaluation is much smaller.

But when I replace my evaluation by PeSTO evaluation I lose about 150 Elo points.

I don't know why this works in Prodeo and not in my engine. I thought that one of the weak parts of my engine was the evaluation, I supposed it was worse than Prodeo and and therefore it seems that it should work easier. So I'm a bit confused.
It's a known fact what works for engine X doesn't have to work for engine Y and vice versa. 4 points comes to mind -

1. I used Pawel's source code for TSCP as found on the TSCP page, pesto.zip.
2. I used the PESTO PST's without the piece values else there will be a conflict with your own piece values.
3. I used the PESTO PST's on top of the PST evaluation in ProDeo, there is some other good stuff in it that I did not want to replace (lose).
4. I noticed the main elo gain came from the evaluation (+45) and little from the search changes, +5 max.
I used the first option. I have tested PeSTO on a TSCP-derived engine (Caligula), so it is even possible to copy the TSCP evaluation directly with copy-paste.

But I understand now, that in ProDeo not all your evaluation has been replaced by PeSTO from TSCP, I understand that you keep things from the evaluation like maybe mobility, king safety ...
Yes, the PESTO evaluation is an add-on to the normal evaluation, solly if that was not clear.
The interesting thing about PeSTO if I'm not confused, is that its author with this evaluation (material + pst) already has 3100 Elo points.
Absolutely amazing, PST only with a good search, conclusion: there must be value is these asymmetrical PST's.
90% of coding is debugging, the other 10% is writing bugs.

dkappe
Posts: 855
Joined: Tue Aug 21, 2018 5:52 pm
Full name: Dietrich Kappe

Re: ProDeo 3.1 - The PESTO version

Post by dkappe » Sun Feb 21, 2021 3:48 am

Did you use the PST values that you shared a few months ago?

User avatar
Rebel
Posts: 5803
Joined: Thu Aug 18, 2011 10:04 am

Re: ProDeo 3.1 - The PESTO version

Post by Rebel » Sun Feb 21, 2021 7:34 am

dkappe wrote:
Sun Feb 21, 2021 3:48 am
Did you use the PST values that you shared a few months ago?
Can't remember that I shared PST's a couple of months ago, I do remember I once did years ago. You will have to refresh my memory :D
90% of coding is debugging, the other 10% is writing bugs.

Post Reply